基于ShuffleNetv2-plus-YOLOX算法的压铸件缺陷检测
Defect Detection of Die Castings Based on ShuffleNetv2-plus-YOLOX Algorithm
蔡振林 1刘斌 1文劲松1
作者信息
- 1. 华南理工大学聚合物成型加工工程教育部重点实验室,广州 510641
- 折叠
摘要
针对压铸件缺陷检测的数据集难以收集、检测效率较低以及工作环境较差等问题,开发了基于YOLOX模型的压铸件缺陷检测软件.用自开发软件的数据增强模块对原始数据集进行增强,解决了压铸件缺陷数据集不充裕的问题;随后将YOLOX算法的Darknet53结构替换为ShuffleNetv2-plus结构,使得利用YOLOX模型检测压铸件缺陷的平均检测精度由原模型的 86.51%提升至89.19%,提升了YOLOX模型识别压铸件缺陷的准确率.
Abstract
Owing to the difficulty in defect detection of die-casting,lower detection efficiency and poor working envi-ronment,a defect detection software for die casting products based on YOLOX model was developed,and the original data set was enhanced by the data enhancement module in the software,solving the problem of the lack of data set.Then,the Darknet53 structure of YOLOX was skillfully replaced by ShuffleNetv2-plus,and finally improved the aver-age detection accuracy of die-casting part defect based on YOLOX model detection from 86.51%of the original model to 89.19%,which greatly improves the recognition precision of die-casting defects by YOLOX model.
关键词
压铸件缺陷/缺陷检测/ShuffleNetv2-plus/YOLOX/数据增强Key words
Die Casting Defect/Defect Detection/ShuffleNetv2-plus/YOLOX/Data Enhancement引用本文复制引用
出版年
2024
NETL
NSTL
维普
万方数据